Extremely accurate ADT7320 temperature sensor

[Langley]

Analog Devices' new ADT7320 temperature sensor is the most accurate digital temp sensor I know of. Typical 25C error of ±0.0017C and max error of ±0.2°C from −10°C to +85. About the only thing better is a Platinum RTD.

Seems to be priced at 4.41USD for 250pcs with 7-wk lead time, so single piece is probably under 10USD, though I didn’t find a distributor with stock yet.
It’s in a tiny 4x4mm 16-lead package and so may be difficult for home brewers to handle, though. Available in both SPI & I2C versions.

Should make for a very nice home lab reference or super accurate weather sensor. I'm planning to interface it to a TI MSP430 Launch Pad as soon as I get my hands on a couple.

http://www.analog.com/en/analog-to-digital-converters/temperature-to-digital-converters/adt7320/products/product.html

Demo video
http://videos.analog.com/video/applications/building-technology/756451946001/ADT7320-and-ADT7420-High-Accuracy-Digital-Temp-Sensors/

[solutions]

7 week lead time?

[Gallymimu]

What are you doing where you could even take advantage of this accuracy?  You are definitely in the realm of a few mA of self heating, puffs of air, thermal gradients and other factors completely blowing accuracy beyond a few tenths of a degree.  Just curious how you are using it and how you are taking advantage of the accuracy because it is hard to do right and I am very curious!

Also, will you be hand soldering that part?  QFNs with slugs on the bottom can be hand soldered but it's tricky.

[Old_but_Alive]

@ solutions.

7 weeks is  a problem ?

just checked the AD site, its status is pre-release, so 7 weeks is to be expected

It is not availabe yet

[FTL]

It is interesting that the device claims a typical temperatures accuracy of 0.0017C with a maximum in the range of 0.2C to 0.65C (depending on temperature and supply voltage). The Max is a LOT larger than the typical, so you can't really count on the typical at all. As well, the 1-bit resolution of the ADC is 0.0078C, so you would never really see a temperature difference of less than 0.01C, so the 0.0017C accuracy sounds pretty meaningless.

[solutions]

I know a lot of the people who work in these places. The industry laid off a lot of competent people and, as a result, quite a few "low cost", "butt-kissing", morons are now left in marketing in these companies.

This sensor is accurate to 0.25C. The fact that the engineers put a 16 bit ADC behind it FOR LINEARIZATION, let someone in MARKETING divide the range by 65,536. It's like your 0.000005% distortion audio amplifier...you can't hear it, but it's "better" to those that are wet behind the ears. As was pointed out here, the mere lag, gradients, convection, and device self-heating will swamp any measurement to the ADC's resolution.

And, yes, 7 weeks lead time is a joke, and so is your project if you are willing to wait around, and then get allocated out of getting your parts by some big dog that needs every one that has been made. The part isn't even out yet (does it even work?) and there are other, equally accurate SPI choices that are in stock and in mainstream use.

[Sideshow Bob]

Typical 25C error of ±0.0017C . Well i doubt that in any real world application. But it may be a good senser then it comes to it

[dotm]

This sensor is accurate to 0.25C. The fact that the engineers put a 16 bit ADC behind it FOR LINEARIZATION, let someone in MARKETING divide the range by 65,536.

that is SO true. you can basically drop the least significant 4 bits. if the measurement error would be gaussian distributed, then you could at least middle some (at least 32 when assuming no variance) measurements, but i think the error of 0.25C will be more like an offset error. No such information in the datasheet.

[Langley]

This sensor is accurate to 0.25C. The fact that the engineers put a 16 bit ADC behind it FOR LINEARIZATION, let someone in MARKETING divide the range by 65,536. It's like your 0.000005% distortion audio amplifier...you can't hear it, but it's "better" to those that are wet behind the ears. As was pointed out here, the mere lag, gradients, convection, and device self-heating will swamp any measurement to the ADC's resolution.

And, yes, 7 weeks lead time is a joke, and so is your project if you are willing to wait around, and then get allocated out of getting your parts by some big dog that needs every one that has been made. The part isn't even out yet (does it even work?) and there are other, equally accurate SPI choices that are in stock and in mainstream use.

I've used the TI (designed by Burr Brown) TMP275 which approaches the ADT7320's claimed accuracy, but I'm unaware of any comparable SPI devices. Which would you suggest?

Yes, self heating can certainly be a problem for this sort of measurement, but low duty cycle sampling helps that a lot.

If the parts actually start shipping I'll try my hand a toaster oven reflow soldering, but if that fails I'll cheat and use the IR re-work oven at work.

Engineering samples exist at least, as the $150 EVAL-ADT7X20EBZ eval board was available to ship tomorrow from the Philippines when I tried a trial order. We'll see how long a couple of sample parts take, though...

[solutions]

0.0017C accuracy for a toaster oven?



There are guys out there successfully using an electric skillet..... bang-bang bimetal thermostat

[Jef Patat]

2^16*0.0078=511.1808 degrees range, while at the same time it states -20 to 105 max

Which just confirms marketing talks 

0.20 accuracy is still nice, but I've never needed it in real world though

Posted on: September 27, 2012, 12:32:39 12:32 - Automerged

0.0017C accuracy for a toaster oven?

Quite right noticing that, better to check soldering info on p6 first!

[dotm]

2^16*0.0078=511.1808 degrees range, while at the same time it states -20 to 105 max

well it says -20 to 105 with garanteed 0.25C accuracy . effective measurement range is -40 to +150
assuming the first bit to be the sign , it would be +/- 255.
try calculating it with 15 bit signed then you will get +/- 127 °C which is below the upper limit, so what's the problem?

[Jef Patat]

I was merely giving another example of misleading info, not of incorrect info, the datasheet even states:

The ADC can theoretically measure a temperature range of 255°C

What's the point of having a better resolution than the accuracy?

Besides the joking, I still think it's a fine chip.

[Magnox]

It probably is a fine chip, but the industry needs to start letting engineers write datasheets again, not clueless marketing morons. Stuff like this does nothing to enhance the companies' or products' reputations.

[dotm]

What's the point of having a better resolution than the accuracy?

Because if you cut down the adc resolution by one bit then you get a dac range that is LOWER than the sensor range.
Got me? This is no marketing gag, it is necessarity.

[solutions]

I was merely giving another example of misleading info, not of incorrect info, the datasheet even states:What's the point of having a better resolution than the accuracy?

Thanks for reading my prior posting.    It's for the linearity calibration table.

They also state that it's only good to 13 bits, so i suspect they may also be doing some dithering to pick up another bit or two out of the measurement noise.  

Posted on: September 28, 2012, 02:26:13 02:26 - Automerged

It probably is a fine chip, but the industry needs to start letting engineers write datasheets again, not clueless marketing morons. Stuff like this does nothing to enhance the companies' or products' reputations.

Trust me - you also don't want datasheets from the engineers. They have no idea what a board or system designer needs to see or what the customers needs guaranteed vs typical, or even what all the apps are or could be.

If the engineers had their way, all datasheet values would be guaranteed or MISSING (vs typical...I have gotten into many a heated discussion just to get a typical value onto a sheet so designers could actually use the part), or the chips would be spec'd into such a corner by perfectionist engineers (they need to be that way to get the chips out the door in a spin or two) that the company would go broke trying to meet all of the parameters in the sheet, or the sheet would be what they can yield, vs what the app needs.

Technically competent marketers/apps engineers used to write sheets. The more experienced ones are driving cabs or greeting people at Walmart now, and the lower priced snot-nosed kids with 3 years of experience are GUESSING what info to put into sheets to sell parts (like this 1E-37 accuracy nonsense), being bullied by politicking marketers that only want a promotion or bonus for themselves, with zero passion for the product line.

Been there, done that, managed it, seen it all (or most of it). Things are picking up a bit, so we might actually get a few Greeters back onto the fold again.

[borberk]

My english is not so good and please explain what means text "Preliminary Technical Data" written on top of each page of Datasheet.

[Sideshow Bob]

My english is not so good and please explain what means text "Preliminary Technical Data" written on top of each page of Datasheet.

It means "for the time being" this is the specifications. But they may change in the final version. Without any notification

[borberk]

It means data are temporary or not final. Thanks.

[solutions]

Preliminary means that the parts have not been run in sufficient quantity and number of material lots (chips are made in batches of wafers, each one a bit different) to generate a sufficient statistical sample size and variance to be able to yield devices to the numbers in the data sheet. We'd normally ask for the fab to run "corners". example 'fast", "slow" and "typical" devices - the corners are process and application specific (some may want Vt for instance instead of Ft, though the two are quite correlated).

That is why the numbers are subject to change, but they are a better guess than what was on the Advanced datasheet, since the data has been verified on the engineering bench in small quantities, usually from the lead lot (first ever) of material.

When manufacturing signs off on the data sheet, it becomes final - they will not stick their necks out because their bonuses and butts are on the line to get parts out that meet the final sheet AND that yield well to ensure decent supply of chips as well as keep the cost the lowest (profit highest).

[alexisnik]

I believe that sensor update frequency is as  important as accuracy (or more important), since these sensors are normally really slow (this one is faster than normal at 4Hz). This goes especially for cases which the application is temperature control, since a thermal system has a time-delay, which makes it difficult to control by adding the delay of the temperature reading... Besides, if the sensor is fast, you can apply some averaging to your readings...

[Jef Patat]

Thanks for reading my prior posting.    It's for the linearity calibration table.

They also state that it's only good to 13 bits, so i suspect they may also be doing some dithering to pick up another bit or two out of the measurement noise. 

Who said I didn't read your post?  If your statement is correct I find the information even more confusing.  That would mean that the resolution isn't even linear, for which I didn't see any evidence in the datasheet (I didn't read every letter though)

From the datasheet:

An internal temperature sensor generates a voltage proportional to absolute temperature; this voltage is compared to an internal voltage reference and input into a precision digital modulator.

I read that what is fed to the ADC is linear and as such I suspect the output of the ADC is linear as well.  I still don't see any reason for the high resolution compared to the accuracy.
Feel free to correct me.

Posted on: October 01, 2012, 11:39:33 11:39 - Automerged

Because if you cut down the adc resolution by one bit then you get a dac range that is LOWER than the sensor range.
Got me? This is no marketing gag, it is necessarity.

Would you?
range form -20 to 105, accuracy 0.25
=> take away the sign we need 105/0.25=420, which is smaller than 2^9 = 512
=> 9 bits + 1 sign bit < 16

[solutions]

You are confusing "linear" as in "straight line" with "linear" as in "analog".

"Proportional" is NOT a straight line by the necessity you infer.

No temperature sensor I know of is a straight line. No sensor I know of is noise free. More bits in the ADC fixes both problems, as I had posted earlier and again you did not read.  More bits doesn't give you information you cannot have by simply throwing bits into an ADC.

You see words, but you are not reading them. You see what you want to see, not what's been written.

[Langley]

Looks like the ADT7320/7420 are finally in full production and distribution. Avnet and DigiKey currently have stock in the U.S. at $7.64 single piece.

Also, here's a nifty $18 eval board for the 7420.
http://www.digilentinc.com/Products/Detail.cfm?Prod=PMOD-TMP2